LogiKEy Methodology for Normative Reasoning

• LogiKEy Methodology is a logic-pluralistic, embedding-driven approach that integrates diverse modal, deontic, and conditional logics into classical higher-order logic.
• It employs a layered architecture with modular shallow embeddings, enabling rapid prototyping and formal verification via advanced HOL theorem provers.
• Its algebraic foundations and logic-parametric design ensure rigorous handling of obligations, permissions, and preference-based reasoning in complex legal and ethical contexts.

LogiKEy Methodology

LogiKEy is a logic-pluralistic, semantical-embedding-driven methodology for ethical and legal reasoning, formal modeling, and the engineering of responsible systems. It enables the rapid prototyping, formal verification, and deployment of normative reasoners by reducing the design of rich logics—including deontic, modal, conditional, and preference-based systems—to modular shallow embeddings in classical higher-order logic (HOL). This approach leverages the tool ecosystem of interactive and automated HOL theorem provers, supports a wide class of object logics, and encompasses a rigorous, algebraic theory of norms, input/output logics, and parametric logic selection. LogiKEy has been developed and validated through both theoretical foundations and extensive tool-supported case studies (Farjami, 2021, Benzmüller et al., 2021, Benzmüller et al., 2020, Farjami et al., 9 Jan 2026, Lawniczak et al., 31 Jan 2025, Benzmüller et al., 2019, Fillottrani et al., 2018).

1. Foundations of the LogiKEy Methodology

The LogiKEy approach, introduced by Benzmüller, Parent, and van der Torre, establishes a plug-in framework for formal ethical and legal reasoning in which domain-specific logics are shallowly embedded into classical higher-order logic (HOL). In a shallow embedding (SSE), each non-classical connective or operator is expressed as an HOL constant or definition. Soundness and completeness (faithfulness) theorems guarantee that reasoning conducted in the target logic corresponds directly to valid reasoning in HOL (Farjami, 2021, Benzmüller et al., 2019).

LogiKEy supports the integration of multiple object logics—deontic, modal, input/output, preference, conditional—without implementing them from scratch, but rather by defining their semantics on top of HOL. This design enables access to the entire ecosystem of proof automation, model-finding, and interactive theorem-proving tools, notably including Isabelle/HOL and external higher-order automated theorem provers such as Leo-III and Sledgehammer, as well as countermodel finders like Nitpick (Farjami, 2021, Benzmüller et al., 2021, Benzmüller et al., 2019).

2. Layered Architecture and Pluralism

LogiKEy operationalizes its methodology through a three-layered architecture:

• L₁ (Logic Layer): Shallow semantic embeddings of object logics into HOL. Each logic (e.g., Standard Deontic Logic KD, Carmo-Jones Dyadic Deontic Logic, input/output logic variants, public announcement logic) is declared as an HOL module, exporting modal, dyadic, preference, or dynamic operators as needed.
• L₂ (Theory Layer): Formalization of general theory components and DSLs or ontologies, such as norm schemas, agency/obligation connectives, context-dependent value structures, or explicit temporal/epistemic modules.
• L₃ (Application Layer): Formal encoding and reasoning over concrete ethico-legal rules, cases, or regulations (e.g., segments of the European AI Act, wild-animal law, GDPR CTD scenarios) (Lawniczak et al., 31 Jan 2025, Benzmüller et al., 2020, Benzmüller et al., 2019).

Pluralism is central: both the choice of object logic at L₁ and the selection of theory components at L₂ are parametric—enabling logic-pluralism, modular experimentation, and context-specific tailoring (Farjami et al., 9 Jan 2026, Benzmüller et al., 2021, Benzmüller et al., 2020).

3. Algebraic Foundations and Input/Output Logics

LogiKEy generalizes normative reasoning via a finitely presented algebraic theory, notably the extension of input/output (I/O) logic over arbitrary Boolean algebras $$\mathcal{B} = \langle B, \wedge, \vee, \neg, 0, 1 \rangle$$ (Farjami, 2021).

• Normative systems are sets $$N \subseteq \operatorname{Ter}(B) \times \operatorname{Ter}(B)$$, where each pair $(a, x)$ represents a “norm” to be implemented: “in context $a$ ensure $x$.”
• I/O operations define outputs $out(N, A) \subseteq B$ for each input $A \subseteq B$, parameterized by closure under Eq-Input, Eq-Output, Strengthen Input (SI), Weaken Output (WO), premise joining (OR), conclusion conjunction (AND), and chaining (T).
• Variants include operations such as:

$out_1(N, A) = Up(N(Up(A)))$

where $Up$ denotes algebraic upward closure.

• Meta-theorems guarantee that algebraic and syntactic (proof-theoretic) closures coincide:

$$out_i(N) = derive_i(N), \quad \forall i \in \{0, R, L, I, II, 1, 2, 3\}$$

• Extensions by AND and cumulative transitivity (CT) are systematically defined and matched with algebraic semantics, supporting deontic aggregation rules (Farjami, 2021).

This structure supports the systematic definition and manipulation of obligations, permissions, and robust treatment of deontic reasoning with CTDs, aggregation, and preference.

4. Embedding, Faithfulness, and Logic-Parametric Tooling

LogiKEy implements shallow semantic embeddings by mapping formulas of the object logic into HOL predicates. Modal, deontic, conditional, and preference operators are introduced via HOL constants, and quantification is inherited from HOL’s simply typed λ-calculus (Benzmüller et al., 2019, Farjami et al., 9 Jan 2026).

• For each target logic, semantic definitions for box/diamond, dyadic obligations, and context-sensitive operators are expressed as HOL terms. For instance, in modal KD:

$$\Box \varphi \equiv \lambda w.\ \forall u.\ r(w,u) \rightarrow \varphi(u)$$

• Faithfulness theorems establish equivalence between object-logic validity and HOL-validity, enabling machine-supported faithfulness proofs:

$$\models^{\mathcal{L}}\ \varphi \iff \models^{\textrm{HOL}}\ \mathrm{vld}\,\llbracket \varphi \rrbracket$$

• Switchable, parametric embeddings facilitate comparative reasoning, as seen in neuro-symbolic NLI settings where the logic layer is passed as a parameter to the pipeline (Farjami et al., 9 Jan 2026).

Automation is achieved by exporting these embeddings to theorem provers and model finders. The LogiKEy workbench in Isabelle/HOL provides out-of-the-box automation for instance theorems, soundness and completeness checks, countermodel generation, and domain-theory verification (Farjami, 2021, Benzmüller et al., 2021, Benzmüller et al., 2020).

5. Generalization to Abstract Logics and DSL Design

LogiKEy abstracts beyond Boolean algebraic settings, allowing I/O logic constructions and embeddings for any abstract Tarskian logic $\mathcal{A} = \langle L, C \rangle$, where $C$ is a Tarskian closure operator and $a \leq b$ iff $b \in C(\{a\})$.

• By replacing algebraic equality and closure with “provable equivalence” and “logical closure,” general I/O operations extend to classic propositional logic, first-order logic, type theory, description logics, and many-modal logics.
• This facilitates the design of multi-logic DSLs and logic-based domain-specific language engineering, as demonstrated in conceptual modeling (evidence-based, ontologically justified lean logic profiles for EER, UML, ORM in conceptual data models) (Fillottrani et al., 2018).

This generalization provides a foundation for hybrid normative systems mixing deontic, temporal, epistemic, preference, and other modalities as needed by the application domain.

6. Conditionality, Preferences, and Dynamic Logics

LogiKEy incorporates algebraic handling of conditional norms and preferences:

• Conditional obligations and permissions are defined via meta-connectives (e.g., $>\circ$ for conditional obligation) and preference Boolean algebras $(B, \vee, \succeq_f)$.
• The framework accommodates both premise-based (local) and global preference orderings. Conditional norms are validated as:
  • $V(\psi) \in out_i(N^V, \{V(\varphi)\})$ for all Boolean algebras and valuations,
  • and, $$\forall V_i \in opt_{\succeq_f}(\varphi):\ V_i(\psi) = 1$$ (Farjami, 2021).
• Dynamic and epistemic logics (e.g., public announcement logic with relativized common knowledge) are realized by extending the SSE approach, introducing explicit domain parameters for updated models and enabling proof automation in dynamic, “model-updating” scenarios (Benzmüller et al., 2021).

This unifies deontic, conditional, preference, and dynamic/epistemic reasoning and enables complex use cases such as the wise men puzzle, dynamic update scenarios, and rich preference- and context-sensitive legal reasoning (Benzmüller et al., 2020).

7. Practical Applications and Tool Support

LogiKEy’s methodology is realized in modular, extensible Isabelle/HOL datasets and workbenches:

• Illustrative examples: Encoding obligations, permissions, CTDs, and preference-based norms in toy and real applications; e.g., GDPR CTD compliance (Benzmüller et al., 2019), property law cases like Pierson v. Post and Conti v. ASPCA (Benzmüller et al., 2020), reasoning over AI regulation (European AI Act) (Lawniczak et al., 31 Jan 2025), and the wise men puzzle (Benzmüller et al., 2021).
• Automated reasoning: Soundness and completeness proofs are automated, performance is empirically validated, and case outcome predictions are checked via Sledgehammer and Nitpick. Countermodels highlight model-theoretic subtleties, like SDL’s failure to consistently validate CTDs versus the DDL/E system’s robustness (Benzmüller et al., 2019, Lawniczak et al., 31 Jan 2025).
• Lessons learned: Test-driven DSL development, reflective equilibrium between logical and domain-level representations, modular logic selection, and iterative extension driven by proof and model feedback are central to the methodology (Benzmüller et al., 2020).
• Toolchain adaptability: Continuous advances in HOL-based ATPs and model-finders directly improve reasoning capability without refactoring the embeddings or theory encodings (Farjami, 2021, Benzmüller et al., 2019).

8. Summary and Significance

LogiKEy establishes a modular, logic-parametric, and mechanized methodology for normative, ethical, and legal reasoning. Its distinguishing features are:

• Faithful, robust shallow semantic embeddings for a wide spectrum of logics in classical HOL.
• Unified algebraic foundations for normative systems, supporting complex closure properties, aggregation, and preference.
• Seamless, tool-integrated pipeline for formalization, verification, and deployment of normative and legal reasoning systems.
• Evidence-based, ontologically aware logic-language design for hybrid and interoperable applications.
• Demonstrated scalability, adaptability, and empirical efficacy in both abstract theoretical investigations and concrete legal, ethical, and epistemic reasoning domains (Farjami, 2021, Benzmüller et al., 2019, Benzmüller et al., 2021, Benzmüller et al., 2020, Lawniczak et al., 31 Jan 2025, Farjami et al., 9 Jan 2026, Fillottrani et al., 2018).

This methodology delivers a foundation for explainable, transparent, and verifiable construction of responsible AI systems and opens a pathway for interdisciplinary research and practical deployment of logic-driven governance, regulatory, and reasoning frameworks.